Visual Attention Mechanism Based Research And Implementation Of Fiber Intrusion Signal Recognition Algorithm

The optical fiber pre-warning system is a new type of security system under the Phase-Sensitive Optical Time-Domain Reflectometer(?-OTDR).It has some advantages,such as high sensitivity,strong anti-interference ability,and real-time monitoring the intrusions in long-distance and subsequently.However,the application environment of fiber optic pre-warning systems is complex and type of intrusion events is multifarious.Identifying intrusion events with high accuracy in different scenarios is a challenge in this field.To the technical difficulty,we propose an algorithm flow based on the visual attention mechanism.First,the signal of the monitoring area is detected and processed to determine the location of the intrusion.Then the fiber signal of the corresponding position is extracted for identification.In addition,we have carried out the research of engineering implementation,and the main accomplishment is shown as follows:First,we summarize the research status of fiber optic pre-warning system,feature extraction methods of fiber signal and visual attention mechanism research status.Next,we summarized the existing methods and implementation bottlenecks.And then,we led to the research ideas and the content of the article.In the end,the composition and principle of the hardware implementation platform constructed by our project are introduced,and the algorithms are verified in field experiments.In fiber intrusion signal preprocessing,according to the birefringence phenomenon of light,the time domain correlation analysis is performed on the intrusion signal.The result of the analysis shows that the false alarm has some relation with the correlation coefficient of the time-domain intrusion signal.Thus,this paper proposes a preprocessing method based on the time-domain correlation coefficient.Thereinto,we optimize two modules,wavelet denoising module and correlation calculation module,which restrict the computational efficiency.The performance of removing false alarm signal is verified with measured data.In the feature extraction of fiber intrusion signals,this paper studies a variety of signal extraction and engineering implementation methods.In time domain,the duty cycle and fundamental frequency are extracted;in the frequency domain,the frequency center feature is extracted.The feature vector constructed of multi-dimensional features has a strong distinction between mechanical and artificial signals.Thereinto,in order to extract the characteristics of artificial signals effectively,this paper also proposes a template matching algorithm based on wavelet threshold to identify artificial signals such as sculpting,trotting,and digging.According to the above feature extraction algorithm,this paper designs an optical pre-warning recognition algorithm flow based on the visual attention mechanism.The idea of visual attention mechanism is the core content of the algorithm flow,and is also implemented on the optical fiber pre-warning system platform.At the same time,the engineering implementation of the recognition algorithm is also an important research content of this article.According to the characteristics of fiber intrusion signal feature extraction,this paper designs feature extraction architecture based on DSP platform,and uses the modularization of ideas to divide functions,mainly including:DSP,host computer communication,storage and computing of DSP data.In the realization of the recognition algorithm,the process is optimized by DSP that improves the efficiency and stability of the system.the fiber intrusion signal preprocessing and identification methods proposed in this paper are verified in experiment.